Multiresolution Fourier Descriptors for Multiresolution Shape Analysis

Abstract

Complex shapes can be effectively analyzed by multiresolution shape descriptors. Compared with wavelet descriptors that are widely used for multiresolution analysis, Fourier descriptors have better invariance properties and higher computational efficiency. We propose a novel multiresolution scheme to generate multiresolution Fourier descriptors for multiresolution analysis: downsampling expansion followed by upsampling reconstruction. Simulation shows that our multiresolution scheme outperforms both wavelet and traditional Fourier descriptors in terms of accuracy and efficiency.